Multi-omics analysis powered by massive data integration

Transcription

1 Multi-omics analysis powered by massive data integration RE(ACT) congress Kristina Hettne, PhD BioSemantics Group, Human Genetics LEIDEN UNIVERSITY MEDICAL CENTER

2 Duchenne muscular dystrophy (DMD) X-linked Mutations in dystrophin - in DMD patients dystrophin is virtually absent; Becker muscular dystrophy patients have 10% to 40% of the normal amount Muscle atrophy, inflammation, fibrosis, fat deposits Incidence 1:5000, costs: $ $ p.p/p.y 2

3 Role of muscle biopsies in clinical trials Muscle biopsies are often requested in interventional clinical trials to show proof of concept that the drug is working e.g. dystrophin restoration after gene therapy, but muscle biopsies are invasive for the patient 3

4 Can bio-signature detection in blood offer an alternative? Identification of blood biomarkers able to monitor disease progression and response to therapy would enable: Better marketing authorization of medicinal products for DMD and muscular dystrophies in general The analysis of blood should support drug developers to show the drug is working (or not working) without taking muscle biopsies 4

5 Aim Gain insights into the effect of dystrophin deficiency on a molecular level identification of novel biomarkers for disease severity and progression 5Insert > Header & footer

6 Research question Identify groups of genes that remain active and drive the pathology 6Insert > Header & footer

7 Dystrophin-deficient mdx mice - Bred at the LUMC - DMD mouse model Nonsense mutation dystrophin - Capable of producing functional analog Utrophin Less severe symptoms 7 Olga Veth 6/6/17

8 Experimental design mdx mice: Samples at 6, 12, 18, 24 and 30 weeks of age Sacrifice at week 30 4 groups, 5-6 mouse per group C57BL or hybrid Pietro Spitali 8Insert > Header & footer

9 Multi-omics data integration, networks and knowledge discovery Problems: Different sizes of data sets Multiple testing burden Solution: Network-based integration Knowledge graphs Rachel Cavill et al. Brief Bioinform, Dec 2015;bib.bbv090 9Insert > Header & footer

10 Approach mdx study 10 Insert > Header & footer

11 Weighted correlation network analysis (WGCNA) 1. Calculate co-expression values 2. Cluster co-expression values 3. Relate modules (lists of genes, metabolites or lipids) to genotype, phenotype, or clinical parameters 11 Insert > Header & footer Langfelder and Horvath, 2008

12 Approach 12 Insert > Header & footer

13 Transcriptomics modules

14 Metabolomics modules

15 Lipidomics modules

16 Calculate significance Does a module have any relation to the disease? Calculate module eigengenes The first principal component of the standardized expression profiles Module eigengenes not always normally distributed Calculate significance by comparing the group means of the module eigengene values using the parametric Kruskal-Wallis test Significance (p < 0.05) indicates a general difference in expression of the module s molecules between one or more group pairs (e.g. wt vs mdx)

17 All metabolomics modules significant, all but one lipidomics module, no transcriptomics modules Example: metabolomics eigengene plots mdx+/+ = mdx/utrn+/+ mdx+/- = mdx/utrn+/- Mohammed Charrout

18 Approach 18 Insert > Header & footer

19 Cross-correlation between module eigengenes reveals multi-omic correlations Module eigengene: the first principal component of the standardized expression profiles Encircled numbers: significant Red line: positive correlation Blue line: negative correlation Mohammed Charrout

20 Cross-correlation filtered for three-way omics Red line: positive correlation Blue line: negative correlation Mohammed Charrout

21 Cross-correlation filtered for three-way omics Turquoise module Highly connected Significant correlations to other omics 0.55 (p=0.01) 0.48 (p=0.04) 0.47 (p=0.04) 0.48 (p=0.04) wt mdx mdx+/+ mdx+/- Red line: positive correlation Blue line: negative correlation mdx+/+ = mdx/utrn+/+ mdx+/- = mdx/utrn+/- Mohammed Charrout

22 Overlap with longtitudinal proteomics in DMD patients * *Proteins changed over time in serum from DMD patients from Spitali et al Under review.

23 Physiologic functional annotation with Euretos Data Sources: Human Protein Atlas Phenotype < > Pathways < > Proteomics < >Metabolomics< > Genetics

24 Euretos physiologic function annotation reveals annotation overlap cytokine signaling process glucose homeostasis biological adaptation to stress cytoprotection negative regulation of circadian rhythm inhibition of circadian rhythm muscle contraction glucose tolerance activation of phosphoinositide 3-kinase cascade endocrine physiology cardiac muscle hypertrophy in response to stress neutrophil degranulation heterophil degranulation fibrinolysis immunity glucose homeostasis energy metabolism heat-shock response nfat pathway uptake physiologic function 24 Insert > Header & footer

25 Functional annotation overlap does not have to mean gene overlap 25 Insert > Header & footer

26 Euretos physiologic function annotation reveals muscle-related processes cytokine signaling process glucose homeostasis biological adaptation to stress cytoprotection negative regulation of circadian rhythm inhibition of circadian rhythm muscle contraction glucose tolerance activation of phosphoinositide 3-kinase cascade endocrine physiology cardiac muscle hypertrophy in response to stress 26 Insert > Header & footer

27 Digging deeper with gene-disease analysis: SCN4A from muscle contraction process Inspired by: Hettne KM. PLoS One Feb. doi: /journal.pone Insert > Header & footer

28 SCN4A is expressed in healthy muscle 28 Insert > Header & footer

29 Summary Disease-related biomolecular modules were found in mouse blood metabolomics and lipidomics data Cross-correlation between module eigengenes revealed multi-omic correlations Significantly correlated turquoise transcriptomics module was related to cytokine production, glucose homeostasis and muscle contraction In conclusion: Multi-omics network analysis reveals a muscular dystrophy-related signature in mouse blood

30 Data management - When published: data in public repositories, scripts on GitHub - Now: - Scripts on LUMC GitLab - Metadata and result data files on FAIR Data Point - WGCNA web tool by Mohammed Charrout creates report with parameters used on GitHub - 6/6/17

31 Next steps Drug repurposing In progress, integrative pipeline to match targets with drugs in place Uptake in the RD-Connect platform Multi-omics analysis pipeline under construction Will incorporate lessons learned from the past years multi-omics analyses: Huntington's Disease (transcriptomics, metabolomics, lipidomics; mouse [1] and human [2]) Duchenne muscular dystrophy (proteomics, transcriptomics, metabolomics, lipidomics; mouse [3] and human [4]) SCA3 (transcriptomics, metabolomics, lipidomics; mouse [5]) Beta-thalassemia (transcriptomics, proteomics, human [6]) [1] Mina E, et al. Orphanet J Rare Dis and in preparation. [2] Mastrokolias A, et al. Metabolomics PMID: [3] Hettne K, et al. in preparation. [4] Spitali P, et al. J Cell Mol Med and under review. [5] Toonen L, et al. under review. [6] Katsantoni E, et al. in preparation.

32 Thank you BioSemantics/Bioinformatics (LUMC) Mohammed Charrout, Eleni Mina, Marco Roos, Mark Thompson, Kees Burger, Rajaram Kaliyaperumal, Barend Mons, Annika Jacobsen Peter A.C. 't Hoen (since Feb 2018 Radboud UMC) DMD exon skip group (LUMC) Pietro Spitali, Annemieke Aartsma-Rus Medical Statistics (LUMC) Roula Tsonaka Profilomic/MedDay Alexandre Seyer 32